Analysis of Multitemporal Remote-Sensing Images for Change Detection: Bayesian Thresholding Approaches

The recent technological developments in remote-sensing sensors and satellites (e.g., the increased spatial and spectral resolutions of sensors, the increased rivisitation time of satellites) offer the possibility of addressing new applications related to environmental monitoring and natural-resource management. In particular, applications connected with the analysis of multitemporal remote-sensing images are becoming more and more important, also in relation to the increased awareness of politicians of the necessity for a regular and efficient control of the environment. This chapter deals with a key issue in multitemporal data analysis, namely, the automatic detection of changes in pairs of images acquired in the same geographical area at different times. In particular, unsupervised change-detection methods (i.e., methods that do not require any ground-truth information for detecting land-cover changes) are considered. Attention is focused on one of the crucial problems that should be solved to develop fully automatic and unsupervised change-detection approaches, i.e., the selection of the decision threshold to be used to distinguish between changed and unchanged areas. Several methods proposed in the literature are described and compared with approaches recently developed by the authors. In addition, some theoretically well-founded thresholding strategies, generally not used for remote-sensing problems, are considered and evaluated also in relation to applications. Examples of results obtained by applying the presented techniques to real multitemporal remote-sensing data sets are reported and discussed.